Hacker News Reader

Comment by drzaiusapelord | Hacker News Reader

> and agree that human judgement should trump an algorithm in what books a library stocks

Why? People are much more biased than machines. If this algorithm is well written and unbiased then we may have a case of someone saving books that speak to their biases. What if it was revealed it was books about young earth, anti-vaccination, promotion of Islamic extremism, and white supremacy? Would you be so welcoming of these rebel librarians?

The nice part of the algorithm is that, in theory, it should just be looking at circulation numbers and culling appropriately. Regardless, this is all a stop-gap measure until we can get every library digital so we don't have to worry about shelf space.

View on HN

There's no such thing as an unbiased algorithm. Machines just amplify the biases of their users. Every choice reflects implicit values and assumptions. More so when you're trying to develop an objective function around the unknown preferences of future library users. What is popular today may not be tomorrow.

And you should really be asking yourself whether circulation numbers are a good proxy for the real value that people derive from the collection. My mother used to borrow ten books at a time and barely read any of them. Building a rigorous utility-maximizing model here is impossible. Human judgement is required.

A library is expected to have a social value beyond just supplying books, this gives it an intrinsic bias by design. And if a library is intended to be comprehensive some books will always be less popular. Librarians are trained to make those kind of disinctions.

I agree that this is certainly a concern, everyone has a bias and we should be aware of it. On the other hand this bias isn't always bad. A librarians bias presumably comes from reading hundreds of books and years of education. If they think an unpopular book is worth saving presumably it is because the book has a high value for the few people who do read it.

It also comes down to trust in the librarian, presumably in the interview process you try to make sure you are not hiring an outspoken Nazi. If you do hire a Nazi librarian they have many other ways of injecting their bias by inviting speakers, recommending books etc. so it is still going to be an issue even if they aren't picking what books you have.

An algorithm that only takes into account popularity and not reviews nor importance of the work to literature or to its technical field contains its own biases.

Says who? It could cross-reference a list of important cultural works and make them exempt, for example. That list can be compiled by a committee instead of a lone guy who may have an agenda.

But the whole point of librarians is to allow that to happen, rather than thinking there's one perfect list of books that everyone should just copy from. For sure you would see lots of bias - which you could then measure and challenge, but which could also turn out to be valuable.

It's odd to me that you can't see the inherent bias in saying that popularity is a proxy for literary value and designing your algorithm around that. Sure, it'll give you an unbiased insight into what's popular, but that assumes that libraries are in the business of catering to the popular taste, when we have a bustling private sector to do that for us already, which takes account of the fact that a great deal of the publishing industry's output is disposable and of only short-term interest. I mean, McDonalds is arguably the world's most popular restaurant chain, but that doesn't mean their food has significant gastronomic value, does it?

What is in the public interest, and what the public is currently interested in, are two wholly different things.

Explicitly favoring recent checkout counts over the other criteria is a bias. This is almost a tautology.

Choosing based on criteria you favor rather than equally valid or possibly more valid criteria others favor is a bias. It does not matter if you encode that bias verbally, in print, or in software. A bias is a bias is a bias.

Reply on news.ycombinator.com